How It Works
The Applied AI Field Guide is structured as an open-source design and verification kit tailored for applied-AI engineers, product owners, workflow managers, service operators, and forward-deployed engineers (FDEs). Its core philosophy centers on starting with real-world work and accepted outcomes rather than abstract model or agent topologies. The guide emphasizes selecting the smallest viable mechanism—whether deterministic software, optimization, classical ML, retrieval, foundation-model calls, bounded agent workflows, or human review—to safely accomplish a task.
The guide operates through a series of structured steps, beginning with understanding the request and workflow, followed by observing and reconciling the work. Each transition between stages requires evidence and accountable decisions, ensuring that model scores, sponsors, or deadlines do not override critical checks for value, authority, safety, ownership, or production readiness. The process is iterative, with each phase building on the previous one to validate working artifacts before they are fully developed.
Central to the guide’s methodology are the 12 Factors of AI Value Engineering and a one-page scorecard. These tools assess outcomes, verifiers, adoption, authority, cost, and proof, providing a framework for testing hypotheses and supporting investment decisions. The guide also includes an executive funding route to align AI initiatives with organizational goals and resource allocation.
The guide is divided into three primary layers: the Five-Minute Guide for mental models and delivery loops, the Handbook for live engagements, and the Engineering Kit for contracts, controls, architecture, evaluations, and operations. These layers are interconnected and designed to be used together, with the capability roadmap helping teams staff workflows effectively. The guide avoids being a standalone framework, instead integrating these components into a cohesive workflow.
Practical application begins with a business flow, where teams accept workflows and value hypotheses before selecting mechanisms. The guide provides runnable examples, such as the invoice practice packet, which simulates real-world scenarios like messy briefs, difficult documents, and review rubrics. These examples compare rules-based approaches with optional model proposals, offering hands-on experience in evaluating retrieval, source support, permissions, freshness, abstention, latency, and cost. The examples are intentionally simplified to illustrate core concepts, with warnings that teaching conveniences should be replaced by target evidence in production environments.
Validation is a critical component of the guide. Artifacts can be validated at different stages—starter or complete—using predefined profiles that check essential fields while maintaining canonical types and closed-object rules. Commands like npm run validate:artifact allow teams to verify workflow charters or other artifacts against these profiles, ensuring consistency and completeness before deployment. The guide also includes a local plugin for managing sources, revisions, decisions, dependencies, and review packets, though it explicitly states that local execution does not grant permission to copy restricted content.
Each step is tied to evidence and accountable decisions, reinforcing the principle that AI systems must be rigorously tested and validated before reaching production. The guide’s documentation is complete as standalone material, with optional skills providing focused routes for teams to explore specific aspects of the workflow without altering the core artifacts.
Enterprise integration and scale reality are emphasized as critical considerations when adapting the guide’s examples to real-world scenarios. The guide acknowledges that simulated services and local state persistence in labs do not equate to production readiness or deployment approval. Teams are advised to replace teaching conveniences with actual evidence and operational constraints before implementing solutions at scale.
The guide’s maintainer, David Ahmann, brings experience as a cloud, data, and applied-AI platform leader with Field CTO background. The project is independent, combining delivery experience, source-linked research, and clearly labeled teaching examples. It is not affiliated with any employer’s guidance or endorsement, ensuring objectivity and a focus on practical, field-tested methodologies.
Use Cases
The Applied AI Field Guide is designed to address a wide range of challenges in applied AI, particularly in scenarios where AI systems must transition from demos to production-ready solutions. For example, in invoice processing, the guide’s invoice practice packet allows teams to test rules-based systems against optional model proposals, identifying failure behaviors and test cases that ensure robustness.
The guide also excels in scenarios requiring safe retry or recovery paths, particularly in enterprise integration contexts. The durable-recovery lab demonstrates how to establish stable operation identities, explicit ambiguity handling, source-of-truth readbacks, and escalation paths.
The guide provides walkthroughs for evaluating change impacts, such as the wrong-amount change walkthrough, which guides teams through source-owned cases, candidate fixes, neighboring regressions, dependencies, and release decisions. This structured approach minimizes the risk of unintended consequences when deploying updates, ensuring that changes are thoroughly vetted before affecting production systems.
The Five-Minute Guide and field engagement tools help teams reframe misaligned projects by identifying representative cases, conflicting claims, safe fallbacks, and decision-makers. This reframing process is essential for realigning AI initiatives with business needs and avoiding costly detours caused by unrealistic expectations or poorly defined objectives.
The guide’s use cases extend to learning by doing, with practical exercises like the invoice practice packet enabling teams to gain hands-on experience. These exercises simulate real-world challenges, such as handling messy briefs or difficult documents, and provide deliberate failures and review results to highlight the limits of evidence.
Industry-specific applications are also supported, though the guide positions its solution portfolio as a design hypothesis rather than a deployable product. Teams can adapt the guide’s methodologies to various sectors, such as finance, healthcare, or logistics, by selecting relevant business-flow patterns and industry profiles. This flexibility ensures that the guide remains applicable across diverse use cases while maintaining a focus on evidence-based decision-making.
Key Features & Capabilities
The Applied AI Field Guide includes several key features designed to streamline the development and deployment of AI systems. One of its standout capabilities is the 12 Factors of AI Value Engineering, a framework that evaluates outcomes, verifiers, adoption, authority, cost, and proof. This scorecard helps teams assess the viability of AI initiatives and align them with organizational goals. The executive funding route further supports this by providing a structured approach to securing investment for AI projects based on their potential value and risk profile.
This kit ensures that teams have the necessary tools to design, build, and operate AI systems in a controlled and reproducible manner.
Runnable examples are a cornerstone of the guide, providing hands-on experience with real-world scenarios. The invoice practice packet, for instance, simulates invoice processing with messy briefs and difficult documents, allowing teams to compare rules-based approaches with model-driven solutions. These examples are designed to be adaptable, with warnings that teaching conveniences should be replaced by production-ready evidence before deployment. The guide also includes labs for retrieval evaluation, durable recovery, and hybrid systems, each addressing specific challenges in AI development.
Validation is a recurring theme in the guide, with tools like npm run validate:artifact enabling teams to check workflow charters and other artifacts against predefined profiles. These profiles ensure consistency and completeness, whether at the starter or complete stage. The guide’s validation process is rigorous, focusing on declared local behavior rather than broader production readiness or deployment approval. This ensures that teams can iteratively refine their solutions while maintaining alignment with business objectives.
The guide’s local plugin adds functionality for managing sources, revisions, decisions, dependencies, and review packets. This plugin enhances continuity and traceability, though it explicitly states that local execution does not grant permission to copy restricted content. Teams are encouraged to keep sensitive or proprietary information within approved source systems, ensuring compliance and security.
Team coverage mapping is another feature introduced in recent updates, providing a workflow-specific overview of roles such as business analysis, product and UX, architecture, data, engineering, evaluation, security, adoption, and service operation. This map describes contributions rather than mandating specific team structures, allowing organizations to tailor their approaches based on available skills and capacity. The guide also extends delivery plans and engagement skills to address decision rights, actual skills, delegates, and coverage gaps, ensuring that teams are adequately prepared to handle AI initiatives.
The guide’s approach to mechanism selection is a defining capability, encouraging teams to compare deterministic software, optimization, classical ML, retrieval, foundation-model calls, bounded agent workflows, and human review. By choosing the smallest viable mechanism, teams can minimize complexity and risk while maximizing the likelihood of success.
The guide’s emphasis on replacing teaching conveniences with target evidence ensures that teams are prepared for the realities of large-scale deployment.
The guide’s optional skills provide focused routes for teams to explore specific aspects of the workflow without altering the core artifacts. These skills are designed to complement the guide’s documentation, offering additional context or tools for teams with particular needs. However, the guide clarifies that skills grant no authority or evidence, reinforcing the importance of accountable decision-making and rigorous validation.
Recent updates to the guide have included enhancements such as workflow-specific team coverage maps, extended delivery plans, and refreshed public navigation and plugin snapshots. The guide’s validation process, including full local suite tests, official plugin validators, and CI checks, guarantees that updates are thoroughly vetted before release, maintaining the guide’s reliability and effectiveness.